It is absolutely critical to verify that the amount of liquid transferred from well-to-well is accurate across the plate, for all steps.
Some assumptions, like habits, can be unusually persistent. One common assumption is that the performance of liquid handling instrumentation – like the electronic drumbeat of a 70s disco dance tune – is dependably predictable and repeatable, beyond even our longest attention span. However, as we demand more output and faster results – while also imposing increased complexity onto our process – we are finding that our instrumentation can miss a step. With more thorough understanding of our pipettes and automated liquid handlers, we can optimize their performance, and hence our process productivity and quality.
In each of these “Perspectives on Instrumentation” columns for Pharmaceutical Processing, I’ve been inviting your questions and appreciate everyone who took time to write. I have been impressed with the increasing quantity of mail – which demonstrates the growing awareness by scientists, process engineers and quality managers of the effects that liquid handling can have on productivity and data integrity. Further, more of you are asking questions relating to robotic liquid handlers as well as to the principles of statistical measurement and uncertainty.
For this final installment, I’ve looked back over all the interesting questions that came to me this past year and selected five of them for publication.
In one of your articles, you used a common example to illustrate the basic concepts of accuracy and precision. You give the example of Tech A, who produced results of 5.3, 4.7 and 5.0 against a reference value of 5.0, and you say that Tech A is accurate but imprecise. I agree on the imprecision but Tech A is only accurate if you look at the average of the three values. So yes, statistically Tech A is accurate but, if one is looking at real outcomes, which is what we really care about in the lab, Tech A’s performance would be unsatisfactory and in my opinion not accurate at all. I am curious about your thoughts on my comments.
Answer 1: You’ve made a very good observation. When good accuracy (based on the arithmetic mean) is combined with poor precision, we get poor individual results. Good quality in laboratory results requires both good accuracy and good precision (as these terms are usually defined in equipment specifications).
Some standards (such as the ISO 5725 series) use a definition of accuracy that is much closer to what you suggest is needed to ensure quality laboratory data. In these standards, accuracy is taken to mean a combination of trueness and precision, where trueness is defined as “the closeness of agreement between the arithmetic mean of a large number of test results and the true or accepted reference value.” This definition for trueness is what I meant when I used the more common term accuracy in the article. I could have been more precise!
The concept of uncertainty is another way to combine inaccuracy and imprecision. Thanks to the international Guide to the Expression of Uncertainty in Analysis (GUM), uncertainty has an accepted mathematical definition and a carefully considered framework for combining inaccuracy (or lack of trueness) and imprecision to produce one number, called the uncertainty, which characterizes the likely range of errors we would expect to see in the analytical results.
Thank you for your thought provoking question.
I use my automated liquid handler to perform serial dilutions in microtiter plates and I want to understand where the biggest potential for errors would be. Please explain.
Answer 2: The three critical areas of error in serial dilutions, as I see it, are: (1) knowing the concentration of starting reagent; (2) accurate buffer volumes and volume transfer for all steps; and (3) proper solution mixing before each transfer. Because concentration is volume dependent, these three sources of error are directly connected to serial dilutions.
The starting reagent concentration, of course, will not affect the systematic serial dilution process but it should be known and understood. By knowing the initial concentration, you will be able to determine the actual concentration levels as the reagent is diluted across the plate. If you think the starting concentration is X, but in reality it is X/2, the concentration levels across the plate will be experimentally different than theory.
It is absolutely critical to verify that the amount of liquid transferred from well-to-well is accurate across the plate, for all steps. Also, the amount of buffer present in the wells should be known, and the same, across all wells – otherwise, the concentration in some wells will be different than expected. Additionally, if the reagents in the wells are not efficiently mixed and therefore not homogeneous before each transfer, then the dilution process would be seriously flawed. Having non-mixed “pockets” of reagent will definitely cause high error propagation across the plate. The experimental results and assumed theoretical concentration levels across the plate will be flawed and the user may have no indication that inefficient mixing is to blame.
As you can see, this is a challenging issue beset with multiple variables.
I use an eight-channel pipette for testing in 96-well microtiter plates and noticed that either the first or last row in each plate was exhibiting poor precision. Thinking that the pipette was malfunctioning, I sent it in for service, and it came back with a passing calibration sticker. But my problem was not solved, and I still had unacceptable precision in the end row. Finally someone in my lab fixed the pipette for me, by cleaning and replacing some parts in the end channel.
I checked the pipette calibration certificate and saw that the company had tested one of the middle channels carefully (multiple replicates) and then tested one replicate for the other seven channels. Why didn’t the service company know that my pipette was broken? Is it standard practice to only check one channel carefully?
Answer 3: I think that you’ve diagnosed the problem correctly, and put your finger on the issue. It seems very likely that an insufficient number of replicates in the calibration process failed to detect your problem, which was imprecision in one channel. When only one replicate is tested on most channels, it becomes more likely that a bad channel will “get lucky” and pass the test by producing one good result.
For a multichannel pipette (or robot for that matter), it is best practice to check each channel as carefully as we would for a single-channel instrument; but this best practice is not necessarily standard practice. The time and expense (not to mention boredom) needed to carefully calibrate multiple channels, one tip at a time, have led many services to take the shortcut you describe. Plus, many labs may be unaware of what sort of service they are paying for, or their calibration budgets may not support the higher cost needed to “do it right.”
Further complicating the situation is that even the most modern written standards do not strictly specify how thoroughly multichannel pipettes should be tested. For example, ISO 8655-6 (published in 2002) states that “each channel should be regarded as a single channel and tested as such.” but also states that “the number of channels tested may be changed to an appropriate number.”
Given the lack of standardization in service offerings, it is important that laboratories understand the details of what their service provider is doing, and make sure that the service is appropriate for their needs.
You’ve recommended that liquids always be equilibrated to room temperature for best accuracy, but my procedures require that I pipette reagents, which are either just thawed from frozen storage or heated to a higher temperature than room temperature. Will this affect the accuracy of the pipetted volume?
Answer 4: When you are using an air-displacement pipette to deliver warmer or cooler samples relative to the pipette and tips, you may introduce significant errors in the delivered volume. Two variables influence the induced error in these cases: sample temperature and target volume.
Regardless of the pipetted volume, pipetting cold samples will lead your pipette to deliver more than the expected volume, while pipetting warm or hot samples will cause the delivery of less than expected volume. Since handheld pipettes are calibrated at 20 oC, the magnitude of the error you may observe will depend on how much the liquid temperature deviates from the calibration temperature (e.g., samples pipetted at 60 oC induce larger errors than samples at 37 oC).
While the sample temperature determines over- or under-delivery, your target volume plays an even more important role. Generally, error increases with decreasing sample volumes. When using an adjustable-volume pipette, it will deliver the most accurate results when used at or close to its nominal volume. Using such a pipette at or close to its minimal volume setting will significantly increase the induced error.
It is always best to fully equilibrate samples and equipment. If this is not feasible, you should be aware that errors exceeding 65% may be introduced, depending on temperature and volumes used.
The other day, while using my pipetting robot, I ran out of pipette tips. Our stockroom didn’t have any more cases of the non-sterile tips I normally use according to my Standard Operating Procedure. However, there were cases of the same tip in a sterilized version. As I was using these sterilized pipette tips, I visually noticed that the tips appeared to retain more of my liquid sample. Once I finished my assay, I saw that my %CV had increased after the change-over to sterilized tips. Does pipette tip sterilization affect the retention rate of my aqueous solutions? If so, does it increase or decrease the amount of liquid retained?
Answer 5: This question has recently been raised several times, and two of my colleagues have just completed a study on the effects of pipette tip sterilization on the frequency and volume of droplets retained on tips following an aspirate and dispense cycle. All of the tips tested had the potential to retain solution. However, the amount of retained droplets did differ significantly between non-sterilized and sterilized pipette tips.
The data suggest that the process of manufacturing pre-sterilized tips increases droplet retention. More drops and bigger drops tend to stick to tips that have been pre-sterilized. The sterilization method affects the tips in a presently undetermined manner such that the solution tends to remain on the pipette tip material at an increased rate (as compared to non-sterilized tips). Therefore, the characteristics and properties of the pipette tips used must not be overlooked when trying to maximize the accuracy and precision of your assay.
When it comes to ensuring liquid handling quality and data integrity, details make the difference. Whether we are mathematically analyzing accuracy and uncertainty, or just trying to find pipette tips that offer the best performance, great care is needed to ensure that our instruments are doing the quality job that we expect.